Why is everyone talking about ChatGPT? Lets’ find out!

Artificial intelligence, Knowledge, Machine Learning

Why is everyone talking about ChatGPT? Lets’ find out!

Introduction

The latest technology trend is ChatGPT, and we haven’t talked about this, so let’s begin. In this blog, we are going to cover what ChatGPT is, how to use it, whether can it code better than you, whether can you play games with it and many other interesting things.

What is ChatGPT?

ChatGPT is the latest model offering from OpenAI, now what is OpenAI?

OpenAI is an AI research and deployment company. With the mission to ensure that artificial general intelligence benefits all of humanity. Check out more about them from the link.

Now let’s get back to the ChatGPT, GPT means generative pre-trained transform, ChatGPT can be called a generative pre-trained transforming style model designed to interact with the users. In short, we can say that it is an interactive model. A dialogue format makes it possible for ChatGPT to answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests. ChatGPT is a sibling model to InstructGPT, which is trained to follow instructions in a prompt and provide a detailed response.

Currently, it is open for User feedback so do check out this awesome AI Model.

I think that is enough of the theory, let’s jump into the practical side.

Hands-on with ChatGPT

We will be using the official web app, provided by OpenAI.

Once you click on this link, it will ask you for a Login or sign up.

Once you log in or sign up, it will ask some generic things such as your name and phone number, phone number verification is there, so make sure to enter the right details.

Once you are done with the necessary fill-ups, the below screen will pop up, with their disclaimers.

So, these are the various disclaimers that you will see after signing in. Just make sure not to share any confidential or sensitive information.

And this is what the final UI looks like-

Let’s see how the ChatGPT works, firstly we will start with the questions given by them only, then we will start playing around with ours.

Firstly, We will drop some easy questions such as what is exascale and quantum computing, and yup it worked.

Now, let’s see can this AI also works with incorrect language and some typos– let me search for – whats the diff between exascale and quantum computing.

And yes it did, here are the results.

So let’s ask the AI, about someone who is not famous, can he fetch the information from the web and display it?

No, it didn’t, it asked us to provide the info.

Let me tell him something about Utkarsh Shukla and then see how it will response

So, it grasps the info that we provided and has shown the output.

Will this info be transferred to new chats or users?

No, it didn’t not even to the new chat from the same user, It’s Interesting.

Can ChatGPT code for you?

Here comes the surprise, the answer is yes.

Let’s begin with a few simple problems such as a program to return the nth Fibonacci number in python.

Boom! it just worked, here is the code for you, and not only code the execution with the results is also there.

Although the code is right, just for the sake of sanity let’s test this in a pycharm editor.

Awesome! it worked.

Can the AI do some more stuff such as making this code more optimized?

Lets’ find out-

Now, that is really interesting, it has provided a better solution.

Lets, ‘s try some more things-

Is ChatGPT a better code than you?

What a question let’s just sum up the above conversation in some points-

The solution was provided in milliseconds
An optimized solution was also provided
Coding instructions were given such as adding comments, they were implemented very neatly.
Code changes were also implemented successfully
Bugs were also answered correctly when there was a bug in the code shared by the ChatGPT once you type the bug it will correct the issue and will provide you with a bug-free solution.
Code instructions were followed throughout the series until you tell the AI to stop. For example, the comments were asked in the Fibonacci program but were followed in other programs as well.
Tried many problems, and very fast solutions were given with proper explanations.

Is GitHub Copilot better than ChatGPT?

Again what a question, Let’s ask ChatGPT about it, 😂 lol.

Satisfied with the answer? No, ok. In short ChatGPT codes for you while Github Copilot codes with you.

Can ChatGPT play games with you?

And again the answer is yes, I played Chess with ChatGPT where I used the move provided by it, and sometimes it resulted in a wrong move, once you let the AI know it will correct it, but it was there n number of times.

I also played rock, paper, and scissors with the AI, where the point I choose was scissors while the AI had chosen rock, so I let the AI know about the wrong answer, and it apologized doesn’t believe me. See the below screenshots.

So, the one issue that I noted here is, I constantly told the AI about the wrong assumption but it keep on repeating the mistakes, about the results between the scissor and the rock.

I think there is some more work required from the OpenAI side on this part of the ChatGPT, as there are a lot of issues arising.

How ChatGPT works?

According to the OpenAI-

We trained the ChatGPT model using Reinforcement Learning from Human Feedback (RLHF), using the same methods as InstructGPT, but with slight differences in the data collection setup. We trained an initial model using supervised fine-tuning: human AI trainers provided conversations in which they played both sides—the user and an AI assistant. We gave the trainers access to model-written suggestions to help them compose their responses. We mixed this new dialogue dataset with the InstructGPT dataset, which we transformed into a dialogue format.

To create a reward model for reinforcement learning, we needed to collect comparison data, which consisted of two or more model responses ranked by quality. To collect this data, we took conversations that AI trainers had with the chatbot. We randomly selected a model-written message, sampled several alternative completions, and had AI trainers rank them. Using these reward models, we can fine-tune the model using Proximal Policy Optimization. We performed several iterations of this process.

ChatGPT is fine-tuned from a model in the GPT-3.5 series, which finished training in early 2022. You can learn more about the 3.5 series here. ChatGPT and GPT 3.5 were trained on an Azure AI supercomputing infrastructure.